Microserver HOWTO, and gearing up for first test farm runs

[Eric S. Raymond]

My two tasks for today:

(1) Only one things is stalling 1.0 release of the HOWTO: figuring out how
to set the GPIO pin on the Odroid from clockmaker.  I will make another pass
at the documentation anf extracting knowledge from the Odroid support people.

(2) Write a script to manage multi-machine test runs on the farm.

The idea is that I think I  ought to be able to say something like this:

$ farmer --start trialtest fe.conf nd.conf ne.conf

to create a test directory named trialtest, copy three ntp.conf files
into it, upload those files to fe, nd, and ne, and restart time service
on all three machines.

Sometime later, I should be able to say

$ farmer --collect trialtest

and copy all the stats files from all machines to per-machine directories
below trialtest.  Then other scripts, not yet written, can do data
reduction and graphing.

The first test run I have planned will use three Pi 3s with Adafruit HATs.
The config files will be identical, all three using GPSD via SHM as a
primary time source, with three differences:

One will use only the GPS.

One will add pool check servers.

One will add the local system clock (driver 1).

What I think I may see is that the drift over short GPS outages is low enough to
make the local clock as effective a check server as the pool.  If that turns
out to be true we can ship a Pi configuration that is really autonomous.
-- 
		<a href="http://www.catb.org/~esr/">Eric S. Raymond</a>